Your documents. Your data. Your control.

Documents are stored in our encrypted database hosted on Turso (SQLite). They are encrypted at rest and in transit. You can delete any document at any time from your account.

No. Your documents are not used to train Google Gemini, Anthropic Claude, or any other AI model; their API terms prohibit it. Text may be briefly cached by the provider (up to about 30 minutes) to make repeat questions on the same document faster, then discarded.

Before your text is sent for AI analysis (questions, summaries, study guides), Recitare automatically scans for and replaces Social Security numbers, UK National Insurance numbers, labelled NHS numbers, phone numbers (US, UK and international formats), email addresses, credit card numbers, IBANs, street addresses, UK postcodes within an address, and dates of birth with [PROTECTED] placeholders. Pattern matching cannot recognise a person's name, which is what Name Protection is for (see the next question). Text sent for speech synthesis is spoken exactly as written, so it is protected by our providers' no-training terms rather than redaction.

Tick “Protect names in this document” when you upload, or switch it on later from the Study tools menu in the reader. A small language model then runs inside your browser and finds the personal names; your document is not sent anywhere for that step. The list is saved with your document, and from then on each name is swapped for a neutral token on our server before a question, summary, study guide or library search goes to our AI providers, and swapped back before the answer reaches you. Until the names have been found, usually a minute or two, Recitare pauses AI features on all your documents rather than run anything unprotected. What it does not cover: read-aloud still speaks the names, because speech synthesis needs the text as written; a Snap sends a picture, and text inside a picture cannot be swapped out, so Recitare asks before the first Snap on a protected document; and your stored document keeps its names so that you can read it. Detection is tuned for English and can miss unusual names, so treat it as a safeguard alongside your ethics approval, not as a substitute for anonymising data your protocol requires you to anonymise. Switching it on later cannot recall anything already sent.

That decision belongs to your ethics approval, not to us. Your ethics committee or data-management plan determines whether participant data may be processed by a cloud service. What we can do is hand you the exact processing chain to take to them: your text is stored encrypted on Turso, processed only by the AI providers named above (Google Gemini, Anthropic Claude, Inworld) under terms that prohibit training on your data, scanned by our automatic redaction before AI analysis (with optional Name Protection, which finds personal names on your own device and hides them from AI analysis), and permanently deletable by you at any time. If your protocol needs more detail, email support@recitare.app before uploading. We are happy to answer your data-protection office's questions directly.

Yes. Every AI interaction is recorded in an activity log with timestamps, model versions, and content hashes. You can review your complete activity history from your account at any time.

Delete your account from Account, then Security. It is deactivated immediately, your subscription is canceled at once, and everything (all documents and their original files, cached audio, summaries, chat history, bookmarks, API tokens, and activity logs) is permanently erased within 30 days. Sign back in within 30 days to cancel and restore everything. Payment records are held by Stripe for legal compliance. One thing we keep: anonymous feature-usage counts (for example, how many times an export button was clicked) with no name, email, documents, or content attached. Once your account is erased, these can no longer be linked to you.

No. We do not sell, share, or provide your data to any third party. The only external services that process your text are the AI providers listed above (Google Gemini, Anthropic Claude, Inworld), under API terms that prohibit training on your data.

As you listen, Recitare quietly keeps a per-paper reading record: which passages you listened to, where you stopped, figures you snapped, and questions you asked, plus an AI-generated capsule (title, one-line summary, key terms). It exists so the app can do useful things for you: resume exactly where you left off, show what each paper was about in your library, and send you an optional next-day recap with a quick recall prompt. It is visible to you in your library, never shared, and never used to train AI models. Recap emails have a one-click unsubscribe. When you delete a document (or your account), its reading record and all its events are permanently erased with it in the same 30-day purge.

Recitare cites specific page numbers (for example, [Page 3]) when generating summaries and study guides. Every claim links back to your source document so you can verify the output against the original text.

Several layers work together. Listening is verbatim: Recitare reads the exact words in your document, so playback cannot invent anything. For answers, summaries, and study guides, the AI works only from the document you uploaded, not the open web, so responses stay grounded in your source. Every specific claim cites its page (for example, [Page 3]) so you can check it against the original, and the AI is instructed to say when something is not in the document rather than guess. Each answer also carries a confidence indicator, computed from grounding signals such as hedging and coverage, that flags responses worth double-checking. No AI is perfect, so we make outputs traceable and checkable rather than ask you to take them on faith.

All data transmitted between your browser and our servers uses TLS 1.2+ encryption. Documents stored in our database are encrypted at rest. Generated audio is cached securely so replays are instant, and is removed when you delete the document.